Performance Optimization on a RAID System: Design and Implementation of a Fast Indexing Table for Disk Caching

隨著處理器和硬碟之間的效能差距不斷地增大，快取記憶體以及磁碟陣列技術被廣泛地應用以縮短此差距，並提升系統輸出入的處理能力。磁碟陣列技術將資料切割成多個區段，同時存放於多顆硬碟上以增加資料存取的效率，所以要維持快取記憶體和多顆硬碟之間資料的對應關係，必須使用更複雜的快取機制。在深入地研究一個實際的磁碟陣列系統及其快取機制之後，我們發現這個系統花費相當多的時間在檢索記憶體上的資料。因此，我們提出一個以快取索引表實作的檢索方法，大幅縮短複雜的快取機制所需的存取時間。經由這個快取索引表，我們可以快速地在快取記憶體上找出所要求的資料，而省下原本繁雜且冗長的檢索運算。我們利用幾種不同的效能度量方法來評估這個磁碟陣列系統，證實快取索引表消耗極少的記憶體空間資源與處理器時間，可以大量地提升系統的輸出入處理能力。As the performance gap between processors and disks continues to increase, cache memory and Redundant Array of Inexpensive Disks (RAID) have been introduced to narrow the gap and increase the throughput of a storage system. While RAID improves I/O performance by striping data across multiple disks, a more complicated caching mechanism is required for a RAID system. Upon studying the caching mechanism in a production storage system, we found that the processor on the system spent a lot of time in indexing the requested disk blocks into the cache memory. Therefore, we propose an indexing method to reduce the access time for the system via a Fast Indexing Table. Finding requested disk blocks via our Fast Indexing Table is far faster than the original indexing method as it is done without complicated RAID indexing operations. We implemented our new indexing method in the storage system and verified its performance through experiments. Our experimental results show that the Fast Indexing Table accelerates the throughput of the storage system significantly with very little increase on the processor and memory usage.口試委員會審定書 i 誌謝 ii 摘要 iii Abstract iv Table of Contents v List of Figures viii List of Tables x Chapter 1 Introduction 1 1.1 Research Contributions 2 1.2 Thesis Organization 3 Chapter 2 Background 4 2.1 RAID Techniques 4 2.2 Caching Techniques 4 Chapter 3 The Target RAID System and Its Caching Mechanism 6 3.1 Architecture of the RAID System 6 3.1.1 A Tour of a Read Request 6 3.1.2 The Data Transfer Module 8 3.1.3 The State Machine 11 3.2 Caching Mechanism of the RAID System 14 3.2.1 Basic Concepts of Disk Arrays 14 3.2.2 Cache Memory Structure 15 3.2.3 SDDB and CCR 18 3.2.4 Put It All Together 20 Chapter 4 Fast Indexing Table (FIT) 22 4.1 Motivation 22 4.2 Concepts of the Fast Indexing Table 23 4.3 Architecture of the Fast Indexing Table 24 4.3.1 Request Format 24 4.3.2 Table Structure 25 4.3.3 Indexing Mechanism 27 4.4 Implementation of the Fast Indexing Table 29 4.4.1 Modifications on the Data Transfer Module 29 4.4.2 Integration of the FIT and the data transfer module 31 Chapter 5 Experiments and Performance Analysis 34 5.1 Experimental Environment 34 5.2 Execution Time Profile Analysis 36 5.2.1 Profiling with the Time Base Counter 37 5.2.2 Execution Time Analysis 37 5.2.3 Analysis for Large Requests 39 5.3 Speedup of the Data Transfer Module 42 5.3.1 The Internal Command Generator Tests 42 5.3.2 Experimental Results 43 5.4 Speedup of the RAID System 45 5.4.1 The Iometer Tests 45 5.4.2 Experimental Results 46 5.4.3 Performance Improvement via Reducing I/O Transactions 48 5.5 Overhead Analysis 50 5.5.1 Spatial Overhead 50 5.5.2 Temporal Overhead 50 Chapter 6 Conclusion and Future Work 52 References 5